AI Search Attribution: Why Analytics Undercounts ChatGPT and Perplexity Traffic

Your analytics undercounts AI search influence in two ways at once. Visits that do arrive from assistants often carry no referrer, landing in direct traffic where nobody looks for them. And the larger effect never becomes a visit at all: the buyer got their answer, formed a shortlist, and either arrived later by brand search or never clicked anything. Standard attribution reads both cases as "AI drives nothing," which is exactly backwards.

Fixing this requires accepting that AI influence splits into a measurable click stream and an unmeasurable answer stream, then instrumenting each with the right tool.

Why the click stream gets mislabeled

Three mechanics scatter AI referrals across your reports:

• Missing referrers. Clicks from assistant apps and secure in-app browsers frequently strip the referrer header, so the session books as Direct.
• Fragmented sources. Traffic that does carry referrers arrives from a zoo of hostnames (chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com) that default analytics groupings bury under referral noise instead of a channel.
• Assistant browsing on the user's behalf. When an assistant fetches your page to answer a question, that fetch is a bot visit with a user waiting behind it, invisible to JavaScript analytics entirely.

The result is systematic: whatever number your analytics shows for AI-driven visits is a floor, not an estimate.

Why the answer stream never clicks

The bigger share of influence happens inside the answer. The published behavioral data is stark: Pew Research found only about 1% of visits to search pages with an AI summary produced a click on a cited source, and SparkToro's clickstream analysis put US Google searches ending without any click at roughly 68%. Seer Interactive measured organic CTR down about 61% where AI Overviews appear.

A buyer who asks an assistant for a shortlist, reads the answer, and later types your name into Google books as branded search. The assistant did the persuading; your last-touch model credits the brand query. Multiply that pattern across a funnel and AI influence hides inside your healthiest-looking channels. The full numbers live in our AI search statistics roundup.

Instrumenting the click stream

Recoverable, with modest work:

1. Build an explicit AI channel. Group the known assistant referrer hostnames into one channel definition so the visits that do carry referrers aggregate somewhere visible.
2. Capture referrals at the edge. Server-side or middleware-level capture sees the Referer header before client scripts and ad blockers interfere, and catches sessions JavaScript analytics drops. Citlyze's crawler tracking install records AI-referred human visits this way, alongside bot activity, per page and per source.
3. Watch assistant fetches as their own signal. Live-fetch agents like ChatGPT-User hitting a page means an assistant used it in a conversation. That is influence with a timestamp, covered in the AI crawler guide.
4. Interrogate your direct traffic. Segment direct sessions landing on deep content pages, the classic signature of stripped-referrer arrivals.

Measuring the answer stream

The unclickable share cannot be attributed from your site. It has to be measured at the source, by sampling the answers themselves:

• Track your prompts. Run the buying questions of your category against the engines on a schedule and record who gets named and cited. This is answer-side visibility, the input that later shows up as branded search and "heard of you" pipeline. The method is our prompt tracking core, and the sampling discipline it requires is covered in why AI answers change.
• Correlate, honestly. Rising mention rates on comparison prompts followed by rising branded search is the causal story worth telling; a survey field ("how did you hear about us?") with an AI option corroborates it cheaply.

FAQ

Why does ChatGPT traffic show as Direct? In-app browsers and privacy behavior strip referrers, so analytics has no source to attribute and defaults to Direct.

Can I see AI answer influence in Google Analytics? Only its click-shaped shadow, and only partially. The answer-side influence requires measuring the answers themselves, outside your site.

Is AI traffic worth attributing given the small click volumes? The clicks are small; the influence is not. Attribution here is less about crediting sessions and more about not strangling the channel that feeds your branded demand.

Count both streams

Citlyze measures the answer stream with scheduled prompt runs and the click stream with edge-level referral capture, in one workspace. See what assistants say about you and who actually arrives from them, starting at citlyze.com/pricing or with a demo.